Investigation of Sediment Entrainment in Brownout Using

High-Speed Particle Image Velocimetry

Bradley Johnson

J. Gordon Leishman

Anish Sydney

Department of Aerospace Engineering

Glenn L. Martin Institute of Technology
University of Maryland, College Park, MD 20742

Nomenclature

Abstract
ap
CD
CT
D
Dmisc
Dp
DStokes
f
F
g
hs
Ip
k
L
LM
Ls
M
Q
R
U, V, W
U p , Vp , Wp
Ur , Vr , Wr
u
ut
V
V rms
Vs
Vtip
W
x, y, z
y0

p
w

To advance the understanding of the brownout phenomenon, the dual-phase flow environment induced by
a rotor hovering above a sediment bed was studied using high-speed flow visualization and particle image velocimetry (PIV). The high frame rate of the camera, combined with advanced particle recognition software, permitted an understanding of the temporal evolution of the
rotor wake in ground effect, simultaneously with the processes of sediment uplift. High-resolution PIV measurements in the surface boundary layer showed large excursions in the ground shear produced by the convecting wake vortices, these excursions being correlated with
localized, intermittent increases in sediment entrainment
rates. Once entrained, significant quantities of sediment
were trapped and vertically transported by the vortexinduced upwash field. Large sediment particles were often
spun out of the vortical flow, and proceeded in a modified saltation trajectory. In particular, the surface and
upwash velocities were shown to strengthen significantly
during the viscous merging of adjacent wake vortices.
This mechanism proved fundamental in defining the concentration of entrained sediment, and also the maximum
height to which sediment could be transported. Particles
reaching sufficient heights were observed to recirculate
into the rotor wake, and convect back towards the ground,
causing sediment ejection through the process of reingestion bombardment. While providing new insight into the
time- and length-scales associated with sediment entrainment by a rotor wake, the observations made here also
bring into question the validity of equilibrium particle flux
models currently being used for brownout simulations.

NDSEG Fellow. E-mail: bjo212@umd.edu

Minta Martin Professor. E-mail: leishman@umd.edu
Minta Martin Intern. E-mail: arbiter@umd.edu
Presented at the 65th Annual National Forum of the American Helicopter Society, Inc., May 2729, 2009, Grapevine, TX.
c
2009
by Johnson et al. Published by the AHS International,
Inc. with permission.

acceleration of sediment particle

drag coefficient of sediment particle
rotor thrust coefficient, = T /2 R4
aerodynamic drag
miscellaneous drag
sediment particle diameter
Stokes drag
empirical function
dust flux
gravitational constant
characteristic saltation height
inter-particle cohesion forces
von Karman's constant (= 0.4)
aerodynamic lift
Magnus lift
characteristic saltation length
aerodynamic moment
saltation flux
radius of blade
Cartesian flow velocities
Cartesian sediment velocities
relative sediment velocities
friction velocity
threshold friction velocity
maximum swirl velocity
rms turbulent velocity
settling velocity
tip speed of blade
sediment particle immersed weight
Cartesian coordinate system
characteristic height
empirical function
fluid density
sediment particle density
wall shear stress
vortex wake age

Introduction

ent rotorcraft, suggesting that there may be certain (perhaps unique) rotor design features that influence the severity and extent of brownout. These features may include
(but are not limited to): rotor disk loading, blade loading, tip speed, number of blades, tip shape, number of
rotors, etc. Assessing the potential of either design or piloting options, however, requires an ability to understand
and predict the complicated fluid dynamics of brownout.
This goal ultimately entails the accurate prediction of the
three-dimensional, unsteady flow field induced by a rotorcraft operating in ground effect, combined with accurate
calculations of the trajectories and non-uniform concentrations of the sediment particles lofted into the rotor flow.
Because multiple phases of matter contribute to the
problem of brownout, at a fundamental level the phenomenon must be considered a multiphase fluid dynamics problem. This multiphase aspect drastically increases
the complexity of measuring, understanding, and modeling brownout because the sediment and carrier fluid are
intrinsically coupled and are capable of transferring momentum and energy to and from each other (i.e., dilute
sediment concentration assumptions may be inadequate).
From a fluid mechanics measurements perspective, there
are clearly many challenging issues involved in understanding such problems, some of which are discussed in
the present article. From a modeling perspective, the issues are equally daunting in that computing the uplift of
the sediment and the subsequent motion of each and every
residual dust particle dispersed in the flow is far beyond
present computing capabilities (Ref. 2).
Despite this hurdle, attempts at simulating brownout
cloud formations have already been undertaken (Refs. 3
6), each simulation invoking various simplifications and
assumptions for calculating the development of the dust
clouds. The most common approach has been to solve
for the induced velocity field at the ground with some
combination of a rotor wake and boundary layer model,
and then to use semi-empirical particle flux and threshold models to represent the initial mass fluxes of sediment being entrained into the rotor flow field. This approach generally requires three types of assumptions: 1)
Those made to simplify the modeling of the rotor flow
field during ground effect operations; 2) Those made to
simplify the modeling of the sediment uplift into the rotor flow field; 3) Those made to simplify the modeling
of the momentum exchange between the multiple phases
of the combined flow (i.e., dilute or one-way coupling).
While most of the assumptions used for rotor wake models (i.e., type 1) have been well studied and validated, the
models being used for entrainment and transport of sediment (i.e., types 2 & 3) are inadequately validated for
brownout problems. Most of these models are also questionable in terms of their underlying assumptions. Consequently, current brownout models must be considered

In recent years, the phenomenon of brownout has become an increasingly significant problem in rotorcraft operations. Brownout is characterized by the formation of
a large, and sometimes blinding, dust cloud that is stirred
up by a rotorcraft when it tries to land or take off from terrain covered with sand or other loose materialsee, for
example, Fig. 1. The intensity of this dust cloud can result in a severe visual obscuration that can cause the pilot
to rapidly lose situational awareness, this problem having
contributed to many helicopter landing and takeoff accidents. Specifically, it has been reported that a significant
number of helicopter accidents have occurred during military operations in the Middle East as a direct result of
pilots suddenly losing situational awareness during inadvertent encounters with brownout.
Within the past few years, new technical efforts have
begun to help reduce the piloting risks posed by brownout
occurrences. Much work has been focused on the development of avionics systems for the pilot by using new sensors and advanced cockpit display technologies (Ref. 1).
While the role of sensor and display technology cannot
be underestimated, the development of special piloting
tactics may still be required to reduce the severity of
brownout, and to control the rapidity at which the dust
cloud develops. However, some these piloting tactics may
not be unique (or even safe), and will depend at least partly
on the type of sediment and non-erodible surface elements
that are present on the ground. The needed tactics to mitigate brownout occurrences may also differ considerably
between different rotorcraft types, i.e., they may be different for single rotor helicopters versus tandem rotor helicopters, helicopters versus tiltrotors, etc.
Ultimately, a more permanent solution to brownout
problems may have to encompass certain design changes
to the rotorcraft itself. For example, based on anecdotal
evidence it is known that intensity and spatial extent of
a developing dust cloud can be very different for differ-

Figure 1: A helicopter encountering brownout conditions during landing.

2

only on a conditional basis, and must be subject to much

further examination as more detailed measurements of the
problem begin to be obtained. Such data must be collected
from several sources, including laboratory tests on subscale rotors, and from field measurements made on actual
helicopters encountering brownout conditions.
The goal of the present work was to help improve the
fundamental understanding of the physics governing the
uplift and transport of sediment in non-uniform, unsteady
flows, characteristic of those produced by rotors in ground
effect. In fact, apart from one previous laboratory experiment (Ref. 7), very little physical understanding has
been obtained about the mechanisms that govern sediment
transport in these types of complex vortical flows. A series of detailed experiments were performed with a small
two-bladed rotor system that was operated in hover above
a ground plane. Different particle species were distributed
on the plane to simulate a sediment bed below the rotor.
Laser sheet illumination and imaging of the flow using
a high-speed digital camera allowed for both dual-phase
flow visualization and time-resolved dual-phase particle
image velocimetry (PIV) measurements of the rotor wake
flow and the sediment uplift processes. The use of advanced particle recognition and tracking algorithms allowed for separate measurements of the flow and sediment
velocities. High-resolution phase-resolved PIV additionally allowed for a study of the developing wall jet and unsteady boundary layer formed by the rotor flow. The results obtained contribute to the understanding of the fluid
dynamics and sediment transport mechanisms that are responsible for the development of brownout conditions.

the surface wind velocity reaches a critical or threshold

value. At this point, particles begin jumping along the
ground in the direction of the flow, a phenomenon called
saltation. Below this threshold velocity, no particle flux
occurs unless an auxiliary force other than the steady wind
disturbs the sediment bed. Therefore, the particle flux
ejected from a sediment bed tends to be a non-linear discontinuous function of flow velocity, the discontinuity existing at the threshold velocity at which the particle flux
originates. The correct identification of the threshold conditions at which particles are mobilized, therefore, plays a
critical role in correctly predicting the onset of any flowdriven particle mobilization process, including the problem of brownout.
Understanding the physics behind this threshold condition requires an examination of the forces acting upon
free particles lying on a flat bed below an external flow
see Fig. 2, which is adapted from Refs. 9 and 10. Acting
downward through the center of gravity is the immersed
weight of the particle W , while the aerodynamic force
(which can be resolved as a lift and drag) is produced by
the action of the boundary layer flow. An aerodynamic
moment, M, may also exist if the center of pressure of the
particle does not align with its geometric center, such as is
typical of non-spherical particles. The inter-particle cohesive forces, IP , are more difficult to define, and comprise
a combination of electrostatic charge, humidity, surface
shape distribution, particle size distribution, and van der
Walls forces (Refs. 911).
Because the sediment bed lies in and/or below a surface boundary layer, the normal approach in modeling the
aerodynamic forces is by using an equivalent wall shear
stress, w . The flow over the bed generally creates a turbulent boundary layer that can be described by a characteristic height, y0 , with a mean velocity U(y) that can be
expressed as
 
y
u
ln
,
(1)
U(y) =
k
y0

Background
Understanding (and eventually predicting) the characteristics of the clouds formed during brownout conditions first
requires a knowledge of sediment transport physics. Because of the difficulty in trying to model individual particle dynamics and inter-particle collisions at the microscale, sediment transport mechanisms tend to be represented by meso-scale models. Most models are based on
the extraction of correlation coefficients from experiments
of specific flows passing over sand and/or soil beds of certain known (or estimated) characteristics (Ref. 8). Of specific interest is the mass flux and concentration of sediment that can be uplifted and introduced into the external
flow, and how to model such processes as a function of the
shearing action exerted by the flow on the sediment bed.

where k is von Karman's constant. The u term is called

the friction velocity, which is proportional to the velocity

Threshold Velocity Criterion

Figure 2: Forces acting on sediment particles at rest
beneath a boundary layer flow.

Bagnold (Ref. 8) observed that the significant mobilization of loose particles on a sediment bed only occurs when
3

gradient, and is defined by

r
u =

w
,

(2)

where w is the surface shear stress and is the density of

the flow. Physically, u is a measure of the mean velocity
gradient in the boundary layer with respect to logarithmic
height off the ground (Ref. 8). The aerodynamic forces
(lift and drag) and the overturning moment can then be
correlated empirically to the size (diameter) of the particle and to the friction velocity (Refs. 9, 10). Therefore, the
higher the friction velocity u (or equivalently, the larger
the shear stress w is) for a given distribution of bed particles, the larger the aerodynamic forces and moments.
Referring again to Fig. 2, the resultant aerodynamic
forces will tend to roll the particles about point P in the
direction of the flow, so causing their movement along
the sediment bed. There will exist some threshold friction
velocity ut at which the aerodynamic forces are exactly
in balance with the combined gravitational and cohesive
forces. Any increase in the friction velocity u beyond ut
will cause the particles to become mobilized. Notice that
an increase in u can result from either an increase in the
external flow velocity above the boundary layer, or by a
change in the boundary layer profile itself. As an outcome
of experimental observations of these initiation processes,
several semi-empirical threshold models have been developed for different types of sediment (Refs. 916).

Figure 3: Spectrum of forces acting on an airborne

sediment particle.
particles follow smooth, ballistic-like trajectories in a
layer just above the sediment bed. This layer is known as
the saltation layer (Refs. 8, 17). Many of the uplifted particles quickly fall back to the ground under gravity, and so
collide with the bed and cause other particles to be set into
motion and so forth. Therefore, saltation can be viewed as
a cascading process, with a perpetual transfer of energy
between the fluid and the particles, and between the free
particles and the sediment bed.
The dynamics of sediment particles within the saltation layer have several dependencies, which will involve
stochastic collision dynamics. Generally, saltating particles are known to rise up nearly vertically from the bed after the threshold conditions are reached, acquiring a vertical velocity Vp,1 and streamwise velocity U p,1 see Fig. 4.
The particles then follow ballistic-like parabolic trajectories with maximum height hs and characteristic length Ls ,
which in practice may range between a few millimeters to
over a meter. Because the particles gain speed and momentum from the external flow during the saltation process, the resultant velocity at impact is larger than the
initial velocity, meaning that the particles impact the bed
with a higher energy state than with which they were first
ejected. Particles too heavy to be lofted into the flow in
saltation roll along the surface in a condition called surface creep.
For a steady, one-dimensional flow passing over a bed
with threshold friction velocity ut , there eventually exists
an equilibrium between the mass flux of particles in saltation above the bed Q and the mean shear stress (or friction
velocity u ) acting on the bed. Therefore, models for Q
are generally derived as a function of the friction velocity
u and the ratio u /ut , i.e., using an equation of the form

Process of Saltation
Once the threshold condition has been met, the sediment
particles follow trajectories based upon the relative magnitude of the forces acting upon themsee Fig. 3. Two
significant parameters are the immersed weight and the
aerodynamic drag, the latter which can be modeled as
a combination of a Stokes drag, DStokes , plus a miscellaneous drag, Dmisc , which results from the (typically)
non-spherical shape of the sediment particles. In addition, Fig. 3 shows the inertial forces produced by acceleration, a p , and the aerodynamic lift which acts in a direction normal to the relative velocity of the particle (i.e., Ur )
with respect to the resultant flow. An additional Magnus
lift force, LM , can be created if the particle has acquired
any spin. Other (but smaller) forces may be present, including apparent mass and Basset forces (i.e., viscous lag)
that result from particle accelerations, and external forces
arising from pressure gradients in the flow (Ref. 9). For
steady one-dimensional winds, the gravitational force is
the dominant force acting on most sediment particles over
50 m in diameter (Ref. 9), which are generally referred
to as sand particles.
When the wind reaches the threshold condition, the
classic mechanism of saltation is produced in which the

  n 
u
Q = f um
,
,
ut

(3)

where m and n are empirically derived correlation

coefficientssee Greeley & Iversen (Ref. 9).
4

You are reading a preview. Would you like to access the full-text?

Access full-text

40 Ramasamy, M., and Leishman, J. G., The Interdependence

of Straining and Viscous Diffusion Effects on Vorticity in Rotor Flow Fields, American Helicopter Society 59th Annual National Forum Phoenix, Arizona, May 68 2003.

27 Ramasamy, M., Leishman, J. G., and Lee, T. E., Flow

Field of a Rotating Wing MAV, Journal of Aircraft, Vol. 44,
(4), pp. 12361244, July 2007.
28 Ramasamy,

M., Johnson, B., Huismann, T., and Leishman,

J. G., , DPIV Measurements of Tip Vortex Characteristics
Using an Improved Aperiodicity Correction, Journal of American Helicopter Society, Vol. 54, (1), January, 2009, pp. 012004
012004.

41 Lee,

T. E., Leishman, J. G., and Ramasamy, M. Fluid Dynamics of Interacting Blade Tip Vortices With a Ground Plane,
Proceedings of the 64th Annual Forum of the American Helicopter Society, Montreal, Canada, April 29May 1, 2008.

29 Ramasamy, M., and Leishman, J. G., Benchmarking Particle Image Velocimetry with Laser Doppler Velocimetry for Rotor Wake Measurements, AIAA Journal, Vol. 45, (11), pp. 2622
- 2633, November 2007.

42 Harvey, J. K., and Perry, F. J.,

Flowfield Produced by Trailing Vortices in the Vicinity of the Ground, AIAA Journal ,
Vol. 9, (8), August 1971, pp. 16591660.
43 Tooby,

R., Scarano, F., Riethmuller, M. L., On Improvement of PIV Image Interrogation Near Stationary Interfaces, Experiments in Fluids, Vol. 45, (4), October, 2008
pp. 557572.

P. F., Wick, G. L., and Isaacs, J. D., The Motion of

a Small Sphere in a Rotating Velocity Field: A Possible Mechanism for Suspending Particles in Turbulence, Journal of Geophysical Research, Vol. 82, (15), May, 1977, pp. 20962100.

31 Leishman, J. G., Measurements of the Aperiodic Wake of a

Hovering Rotor, Experiments in Fluids, Vol. 25, (4), September,
1998, pp. 352361.

44 Nielsen, P., On the Motion of Suspended Sand Particles,

Journal of Geophysical Research, Vol. 89, (C1), January, 1984,
pp. 616626.

30 Theunissen,

32 Bhagwat,

M., and Leishman, J.G., Stability Analysis of

Helicopter Rotor Wakes in Axial Flight, Journal of the American Helicopter Society, Vol. 45, (3), July, 2000, pp. 165178.

45 Nielsen, P., Turbulence Effects on the Settling of Suspended Particles, Journal of Sedimentary Petrology, Vol. 63,
(5), September, 1993, pp. 835838.

33 Tangler, J. L., Wohlfeld, R. M., and Miley, S. J., An Experimental Investigation of Vortex Stability, Tip Shapes, Compressibility, and Noise for Hovering Model Rotors, NASA CR-2305,
1973.

46 Shao, Y., and Raupach, M. R., The Overshoot and Equilibrium of Saltation, Journal of Geophysical Research, Vol. 97,
(D18), pp. 20,55920,564.
47 Wereley, S. T., Gui, L., and Meinhart, C. D., Advanced
Algorithms for Microscale Particle Image Velocimetry, AIAA
Journal, Vol. 40, (6), pp. 10471055, June, 2002.

34 Kiger,

K., and Pan, C., PIV Technique for the Simultaneous Measurement of Dilute Two-Phase Flows, Journal of
Fluids Engineering, Vol. 122, December, 2000, pp 811818.

48 Westerweel, J., Dabiri, D., and Gharib, M., The Effect of

a Discrete Window Offset on the Accuracy of Cross-correlation
Analysis of Digital PIV Recordings, Experiments in Fluids,
Vol. 23, 1997, pp. 2028.

35 Anderson, S., and Longmire, E.,

Interpretation of Autocorrelation PIV Measurements in Complex Particle-Laden Flows,

Experiments in Fluids, Vol. 20, (4), pp. 314317.
36 Paris,

A., and Eaton, J., Measuring Velocity Gradients in

a Particle-Laden Channel Flow Proceedings of the Third International Workshop on Particle Image Velocimetry, September
1618, 2000, Santa Barbara, CA, pp. 513518.

49 Ronneberger,

O., Raffal, M., and Kompenhans, J., Advanced Evaluation Algorithm for Standard and Dual Plane Particle Image Velocimetry, Proceedings of the 9th International
Symposium on Applications of Laser Techniques to Fluid Mechanics, Lisbon, Portugal, July 1316, 1998.

37 Gui,

L., Lindken, R., and Merzkirch, W., Phase-Separated

PIV Measurements of the Flow Around Systems of Bubbles Rising in Water, Proceedings of the ASME Fluids Engineering
Division Summer Conference, June 2226, 1997, Vancouver,
Canada.

50 Usera, G., Vernet, A., and Ferre, J. A., Considerations

and Improvements of the Analyzing Algorithms Used for Time
Resolved PIV of Wall Bounded Flows, Proceedings of the 12th
International Symposium on Applications of Laser Techniques
to Fluid Mechanics, Lisbon, Portugal, July 1115, 2004.

38 Lindken,

R., and Merzkirch, W., A Novel PIV Technique

for Measurements in Multi-Phase Flows and its Application to
Two-Phase Bubbly Flows, 4th International Symposium on Particle Image Velocimetry, G'ottingen, Germany, September 17
19, 2001.

51 Scarano, F.,

Iterative Image Deformation Methods in PIV,

Measurement Science and Technology, Vol. 13, 2002, pp. R1
R19.

39 Ananthan,

S., Leishman, J. G., and Ramasamy, M., The

Role of Filament Stretching in the Free-Vortex Modeling of Rotor Wakes, American Helicopter Society 58th Annual National
Forum, Montreal, Canada, June 1113, 2002.

52 Butterfield,

G. R., Transitional Behaviour of Saltation:

Wind Tunnel Observations of Unsteady Winds, Journal of Arid
Environments, Vol. 39, (3), July 1998, pp. 377394.

21

Y pixel displacement

Appendix A
PIV Measurements Near Interfaces
The challenges in establishing reliable cross-correlations
of PIV images made near an interface are summarized in
Ref. 30 and include: 1) The presence of non-uniform laser
reflections; 2) Signal truncation; 3) A high wall-normal
velocity gradient; 4) Vector relocation when correlation
windows overlap the interface.

16
8
0
-8

-16
-16

-8

8 16 -16 -8 0
X pixel displacement

16

Techniques to Reduce Laser Reflections

Figure A.2: Example of a correlation map (left), and

one contaminated by laser reflections (right).

Because PIV cross-correlation procedures match patterns

in pixel intensities to determine particle displacements,
the presence of non-uniform intensity patterns from laser
reflections can adversely contaminate the correlation process, resulting in poor signal-to-noise ratios and erroneous
velocity estimates. For example, Fig. A.1(a) shows a raw
PIV image sample that overlaps the interface region. In
this case, the interface is contaminated with high intensity
pixels of much higher greyscale values than those from the
tracer particles. When standard cross correlation methods
are used, each high intensity pixel at the interface will correlate well with every other high intensity pixel.
Consequently, a completely biased correlation map will
be produced, with correlation peaks running parallel to the
direction of the interfacesee Fig. A.2. Instead of providing one distinct correlation peak at the true particle displacement, this effect leads to spurious velocity vectors.

Figure A.3: Improved experimental technique and image processing allowed the flow to be successfully measured well into the boundary layer region.
Even in regions where the reflections are less intense
see Fig. A.1(b)the flow measurements made there will
still be biased toward zero velocity. This is because crosscorrelation algorithms cannot differentiate between the
signals produced from tracer particles or from laser reflections. Because the pixel regions affected by laser reflections generally do not change much from the first PIV
frame to the second, they are essentially registered as
tracer particles with zero displacements.
To reduce the intensity of laser reflections in the present
work, a Rhodamine 6G fluorescent paint was applied
to the ground plane. This paint shifted the wavelength
of the reected laser light from 532 nm to 590 nm. A
532 nm notch filter was then placed in front of the camera
such that the high intensity reflected light (now mostly at
590 nm) was filtered out. A second approach used to reduce surface reections was to align the laser sheet so that
the incident light rays were more parallel to the interface.
After acquisition, the images were pre-processed with a
background subtraction techniquesee Ref. 47. A binary
mask image was created in which each pixel was assigned
a 0 or 1 based on its spatial location; all pixels in the
flow area were given a value of 1, and all pixels in the
interface were given a value of 0. This binary mask was

Improved Experimental
Technique
Improved
Experimental

Technique
(a) Unaltered experiment
Improved Experimental
Technique

Background Subtraction
and Masking
(b) Improvement through experimental
techniques
Background
Subtraction

and Masking
Background Subtraction
and Masking

(c) Improvement through background subtraction and masking

Figure A.1: Improvement in near-ground PIV interrogation through experimental and image preprocessing techniques.
22

then multiplied with each instantaneous PIV image, such

that all pixels in the interface ended up with zero intensity.
The overall improvement in the raw images after Rhodamine treatment, background subtraction, and binary
mask generation is shown in Fig. A.1(c). These techniques allowed velocity measurements to be made approximately 2 mm closer to the surface compared to what
was possible by not using these methodssee the velocity profiles in Fig. A.3. The differences in the velocities
away from the wall arise because the two instantaneous
measurements shown are not temporally correlated, and
not because of any artificial contamination of the raw images from the implementation of these methods.

dU1
dy

dU2
dy
Increase in
window
aspect ratio

Boundary
layer flow
Interface

dU2 dU1
<
dy
dy

Figure A.4: Increasing the aspect ratio of the PIV interrogation window reduces velocity gradient bias errors in the wall-normal direction.

Image Processing

Flow

To obtain the flow velocities from two PIV image pairs,

the image must be divided into thousands of small interrogation windows. Intensity patterns in corresponding correlation windows are then matched to measure the average
particle displacements within each set of windows.
In the present work, challenges arise in determining the
optimal cross-correlation algorithms and grids to measure
the particle displacements. Specifically, the flows generated by a rotor operating near the ground interface produces high shear within the vortex cores and also inside
the boundary layer that develops on the interface.
To help improve the measurement resolution and fidelity in these two regions, two separate processing grids
were used for each PIV image pair: one for the boundary
layer flow, and one for the remainder of the flow where
the dominant coherent structures were the wake vortices.
The first processing grid was used to improve the fidelity of the flow measurements near the ground, and was
based on suggestions given by Theunissen et al. (Ref. 30).
In the boundary layer, the largest gradient exists in the
wall-normal direction, so measurements with good spatial resolution are needed to resolve these gradients. A
simple approach is to decrease the size of the individual
correlation windows, and so to derive a larger number of
independent measurement points. This process, however,
limits the number of tracer particles present in each window, and may deteriorate the cross-correlation procedure.
To compensate for this effect, the aspect ratio of the
interrogation windows can be increased such that their
dimensions parallel to the interface are extended, while
their dimensions normal to the interface are reducedsee
Fig. A.4. This allows the boundary layer to be sampled
at a higher spatial resolution in the wall-normal direction,
while still maintaining a sufficient number of tracer particles in the individual windows. Additionally, correlation
windows with the smaller wall-normal dimensions will
decrease the velocity gradient bias errors. This is because
the variations of individual tracer particle displacements

h-hov

Correct
h/2

hov

Incorrect
(default)

h/2

Interface

Figure A.5: Vector relocation is required whenever the

interrogation windows overlap the interface.
within the windows decrease with smaller windows. Figure A.4 shows this issue for both standard square interrogation windows, and for the deformed rectangular windows. Because the dimension of the rectangular windows
in the wall-normal direction has been decreased, the discrepancies in the flow velocity (shown by the green velocity vectors) from the top of the window to the bottom
window are also decreased.
The final dimension of the windows chosen for the
boundary layer grid was 48-by-8 pixels. This allowed for
over 97% of the vectors to pass with a signal-to-noise ratio
of 1.5, while also giving the maximum spatial resolution
in the wall-normal direction. The cross-correlation was
then performed using a two-step recursive technique in
which the integer pixel displacement (m, n) was estimated
in the first processing pass. The correlation windows were
then offset about their geometric center by m/2 and n/2
before starting a second processing pass. This two-pass
technique has been shown to significantly improve the
signal-to-noise ratio of the measurements (Ref. 48). This
grid was extended to 1.5 mm above the surface (below
which the most substantial values of dU/dy were measured), and then combined with a deformation grid above
for the remaining flow, which is optimal for velocity measurements within the wake vortices.
Even if the effects of non-uniform laser reflections are
removed by experimental techniques or by suitable pre23

processing, a signal truncation issue still exists at the

interface. As discussed by Theunissen & Ronneberger
(Refs. 30, 49), this discontinuity in pixel intensity is nonoptimal for preferred Fast-Fourier Transform (FFT) processing techniques because the signal resulting from the
image pattern is no longer periodic. Additionally, when
the correlation window overlaps the interface, the final
vector placement is spatially incorrect. By default, PIV
processing algorithms generally place vectors at the center of the interrogation window. For a window overlapping the interface, however, the center of the correlation
window will not coincide with the center of the truncated
seeded region (see Fig. A.5). Therefore, any vector estimated from a correlation window with overlap must be
mathematically adjusted such that its final location coincides with the center of the seeded region (Refs. 30, 50).
To eliminate such problems, the grid was defined such that
the bottom of the first correlation windows were anchored
at the interface, and arranged such that no windows mutually overlapped this interface.

serves as the first iteration. Once the mean displacement

of that region is estimated, the interrogation window of the
displaced image is displaced by integer pixel values for
better correlation in the second iteration. This third iteration starts by moving the interrogation window of the displaced image by sub-pixel values based on the displacement estimated from second iteration. Following this,
the interrogation window is sheared twice (for integer and
sub-pixel values) based on the velocity magnitudes from
the neighboring nodes before performing fourth and fifth
iteration, respectively.
Once the velocity is estimated after these five iterations,
the window is split into four equal windows. These windows are moved by the average displacement estimated
from the final iteration before starting the first iteration
at this resolution. This procedure can be continued until
the resolution required to resolve the flow field is reached.
The second interrogation window is deformed until the
particles remain at the same location after correlation.

Appendix B Deformation Grid

Algorithm
For TR-PIV processing in general, and for PR-PIV processing above the boundary layer region, the images
were processed with a multi-step deformation grid crosscorrelation procedure (Ref. 51), as also described in
Ref. 29. This procedure is optimized to measure vortical
flows and other flows with steep velocity gradients.
The first step in the process is similar to that of a simple
recursive technique, however, at the second step the second window is deformed (i.e., it is both sheared and translated instead of just the simple translation)see Fig. B.1.
The procedure starts with the correlation of an interrogation window of a defined size (say, 64-by-64), which
64 X 64
First window

64 X 64

R0

64 X 64

R0

Second window

Translation

Shear

First window
32 X 32

R2

R1/2

32 X 32

- R1/2

Second window

R1

64 X 64

Figure B.1: Schematic showing the basis of the deformation grid technique.
24